dma-buf: allow reserving more than one shared fence slot

Let's support simultaneous submissions to multiple engines.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Link: https://patchwork.kernel.org/patch/10626149/

drivers/dma-buf/reservation.c

@@ -56,9 +56,10 @@ const char reservation_seqcount_string[] = "reservation_seqcount";
 EXPORT_SYMBOL(reservation_seqcount_string);
 
 /**
- * reservation_object_reserve_shared - Reserve space to add a shared
- * fence to a reservation_object.
+ * reservation_object_reserve_shared - Reserve space to add shared fences to
+ * a reservation_object.
  * @obj: reservation object
+ * @num_fences: number of fences we want to add
  *
  * Should be called before reservation_object_add_shared_fence().  Must
  * be called with obj->lock held.
@@ -66,7 +67,8 @@ EXPORT_SYMBOL(reservation_seqcount_string);
  * RETURNS
  * Zero for success, or -errno
  */
-int reservation_object_reserve_shared(struct reservation_object *obj)
+int reservation_object_reserve_shared(struct reservation_object *obj,
+				      unsigned int num_fences)
 {
 	struct reservation_object_list *old, *new;
 	unsigned int i, j, k, max;
@@ -74,10 +76,11 @@ int reservation_object_reserve_shared(struct reservation_object *obj)
 	old = reservation_object_get_list(obj);
 
 	if (old && old->shared_max) {
-		if (old->shared_count < old->shared_max)
+		if ((old->shared_count + num_fences) <= old->shared_max)
 			return 0;
 		else
-			max = old->shared_max * 2;
+			max = max(old->shared_count + num_fences,
+				  old->shared_max * 2);
 	} else {
 		max = 4;
 	}

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -955,7 +955,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 	if (r)
 		return r;
 
-	r = reservation_object_reserve_shared(vm->root.base.bo->tbo.resv);
+	r = reservation_object_reserve_shared(vm->root.base.bo->tbo.resv, 1);
 	if (r)
 		return r;
 

drivers/gpu/drm/amd/amdgpu/amdgpu_object.c

@@ -640,7 +640,7 @@ int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
 	bo_addr = amdgpu_bo_gpu_offset(bo);
 	shadow_addr = amdgpu_bo_gpu_offset(bo->shadow);
 
-	r = reservation_object_reserve_shared(bo->tbo.resv);
+	r = reservation_object_reserve_shared(bo->tbo.resv, 1);
 	if (r)
 		goto err;
 

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

@@ -772,7 +772,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 
 	ring = container_of(vm->entity.rq->sched, struct amdgpu_ring, sched);
 
-	r = reservation_object_reserve_shared(bo->tbo.resv);
+	r = reservation_object_reserve_shared(bo->tbo.resv, 1);
 	if (r)
 		return r;
 
@@ -1839,7 +1839,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	if (r)
 		goto error_free;
 
-	r = reservation_object_reserve_shared(vm->root.base.bo->tbo.resv);
+	r = reservation_object_reserve_shared(vm->root.base.bo->tbo.resv, 1);
 	if (r)
 		goto error_free;
 

drivers/gpu/drm/etnaviv/etnaviv_gem_submit.c

@@ -179,7 +179,7 @@ static int submit_fence_sync(struct etnaviv_gem_submit *submit)
 		struct reservation_object *robj = bo->obj->resv;
 
 		if (!(bo->flags & ETNA_SUBMIT_BO_WRITE)) {
-			ret = reservation_object_reserve_shared(robj);
+			ret = reservation_object_reserve_shared(robj, 1);
 			if (ret)
 				return ret;
 		}

drivers/gpu/drm/i915/i915_vma.c

@@ -892,7 +892,7 @@ static void export_fence(struct i915_vma *vma,
 	reservation_object_lock(resv, NULL);
 	if (flags & EXEC_OBJECT_WRITE)
 		reservation_object_add_excl_fence(resv, &rq->fence);
-	else if (reservation_object_reserve_shared(resv) == 0)
+	else if (reservation_object_reserve_shared(resv, 1) == 0)
 		reservation_object_add_shared_fence(resv, &rq->fence);
 	reservation_object_unlock(resv);
 }

drivers/gpu/drm/msm/msm_gem_submit.c

@@ -241,7 +241,8 @@ static int submit_fence_sync(struct msm_gem_submit *submit, bool no_implicit)
 			 * strange place to call it.  OTOH this is a
 			 * convenient can-fail point to hook it in.
 			 */
-			ret = reservation_object_reserve_shared(msm_obj->resv);
+			ret = reservation_object_reserve_shared(msm_obj->resv,
+								1);
 			if (ret)
 				return ret;
 		}

drivers/gpu/drm/nouveau/nouveau_fence.c

@@ -341,7 +341,7 @@ nouveau_fence_sync(struct nouveau_bo *nvbo, struct nouveau_channel *chan, bool e
 	int ret = 0, i;
 
 	if (!exclusive) {
-		ret = reservation_object_reserve_shared(resv);
+		ret = reservation_object_reserve_shared(resv, 1);
 
 		if (ret)
 			return ret;

drivers/gpu/drm/qxl/qxl_release.c

@@ -234,7 +234,7 @@ static int qxl_release_validate_bo(struct qxl_bo *bo)
 			return ret;
 	}
 
-	ret = reservation_object_reserve_shared(bo->tbo.resv);
+	ret = reservation_object_reserve_shared(bo->tbo.resv, 1);
 	if (ret)
 		return ret;
 

drivers/gpu/drm/radeon/radeon_vm.c

@@ -831,7 +831,7 @@ static int radeon_vm_update_ptes(struct radeon_device *rdev,
 		int r;
 
 		radeon_sync_resv(rdev, &ib->sync, pt->tbo.resv, true);
-		r = reservation_object_reserve_shared(pt->tbo.resv);
+		r = reservation_object_reserve_shared(pt->tbo.resv, 1);
 		if (r)
 			return r;
 

drivers/gpu/drm/ttm/ttm_bo.c

@@ -872,7 +872,7 @@ static int ttm_bo_add_move_fence(struct ttm_buffer_object *bo,
 	if (fence) {
 		reservation_object_add_shared_fence(bo->resv, fence);
 
-		ret = reservation_object_reserve_shared(bo->resv);
+		ret = reservation_object_reserve_shared(bo->resv, 1);
 		if (unlikely(ret))
 			return ret;
 
@@ -977,7 +977,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,
 	bool has_erestartsys = false;
 	int i, ret;
 
-	ret = reservation_object_reserve_shared(bo->resv);
+	ret = reservation_object_reserve_shared(bo->resv, 1);
 	if (unlikely(ret))
 		return ret;
 

drivers/gpu/drm/ttm/ttm_execbuf_util.c

@@ -129,7 +129,7 @@ int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket,
 			if (!entry->shared)
 				continue;
 
-			ret = reservation_object_reserve_shared(bo->resv);
+			ret = reservation_object_reserve_shared(bo->resv, 1);
 			if (!ret)
 				continue;
 		}
@@ -151,7 +151,7 @@ int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket,
 		}
 
 		if (!ret && entry->shared)
-			ret = reservation_object_reserve_shared(bo->resv);
+			ret = reservation_object_reserve_shared(bo->resv, 1);
 
 		if (unlikely(ret != 0)) {
 			if (ret == -EINTR)

drivers/gpu/drm/v3d/v3d_gem.c

@@ -305,7 +305,7 @@ v3d_lock_bo_reservations(struct drm_device *dev,
 	for (i = 0; i < exec->bo_count; i++) {
 		bo = to_v3d_bo(&exec->bo[i]->base);
 
-		ret = reservation_object_reserve_shared(bo->resv);
+		ret = reservation_object_reserve_shared(bo->resv, 1);
 		if (ret) {
 			v3d_unlock_bo_reservations(dev, exec, acquire_ctx);
 			return ret;

drivers/gpu/drm/vc4/vc4_gem.c

@@ -635,7 +635,7 @@ vc4_lock_bo_reservations(struct drm_device *dev,
 	for (i = 0; i < exec->bo_count; i++) {
 		bo = to_vc4_bo(&exec->bo[i]->base);
 
-		ret = reservation_object_reserve_shared(bo->resv);
+		ret = reservation_object_reserve_shared(bo->resv, 1);
 		if (ret) {
 			vc4_unlock_bo_reservations(dev, exec, acquire_ctx);
 			return ret;

drivers/gpu/drm/vgem/vgem_fence.c

@@ -180,7 +180,7 @@ int vgem_fence_attach_ioctl(struct drm_device *dev,
 	reservation_object_lock(resv, NULL);
 	if (arg->flags & VGEM_FENCE_WRITE)
 		reservation_object_add_excl_fence(resv, fence);
-	else if ((ret = reservation_object_reserve_shared(resv)) == 0)
+	else if ((ret = reservation_object_reserve_shared(resv, 1)) == 0)
 		reservation_object_add_shared_fence(resv, fence);
 	reservation_object_unlock(resv);
 

include/linux/reservation.h

@@ -261,7 +261,8 @@ reservation_object_get_excl_rcu(struct reservation_object *obj)
 	return fence;
 }
 
-int reservation_object_reserve_shared(struct reservation_object *obj);
+int reservation_object_reserve_shared(struct reservation_object *obj,
+				      unsigned int num_fences);
 void reservation_object_add_shared_fence(struct reservation_object *obj,
 					 struct dma_fence *fence);